Screening for Neonatal Jaundice With a Mobile Health Device: a Validation Study in Oaxaca, Mexico.
Screening for Neonatal Jaundice With a Mobile Health Device: a Validation Study in Oaxaca, Mexico.
The performance and accuracy of the smartphone application (app) Picterus Jaundice Pro (PicterusJP) will be validated by a cross-sectional study conducted in low resource settings in Mexico. Physics-based and machine learning-based models will be used to analyze the images obtained with the app and its performance will be compared.
A cross-sectional prospective study is taken place at Hospital General de Zona 1 in the city of Oaxaca, Mexico.
Participants are recruited from the maternity ward and emergency room and data collection is conducted by one experienced pediatrician and one paediatrics' resident. After informed consent is obtained, background data including time and date or birth, birth weight, gestational age, parent's skin type according to Fitzpatrick scale and newborn's skin color according to Neomar's scale is being collected. Visual assessment is performed by the researchers using Kramer's scale. To obtain the bilirubin measurements with the smartphone app Picterus Jaundice Pro (Picterus JP), the calibration card is placed on the newborn's chest and a set of six images, 3 with and 3 without flash, are captured. The smartphones used to capture the images are Samsung A23, iPhone 11 and iPhone 15 Pro.
Total serum bilirubin (TSB) levels are measured with a blood sample obtained within one hour of capturing the images and analyzed at the hospital laboratory facilities.
Demographic and clinical characteristics will be analyzed by descriptive statistic.
Bilirubin measurements correlation between Picterus JP using both models and TSB will be determined by Pearson's correlation coefficient. Systematic over or underestimation of bilirubin levels will be evaluated using Bland-Altman plots. Accuracy measurements will be determined using receiver operating characteristics (ROC) and area under the curve (AUC) analysis. To compare the correlations from both models, Fisher's r to z transformation approach will be applied
Inclusion Criteria:
Exclusion Criteria: